Design of an Integrated Model Using Blockchain-Supported Federated Deep Reinforcement Learning and Convolutional Recurrent Federated Learning for IoV Networks

Abstract

With the increasing penetration of IoV, there is an emerging need for new, more advanced secure, efficient, and scalable methods for resource management. Approaches such as centralized learning have limited applicability in the highly dynamic and mobile environment, citing problems such as privacy breach, latency, and overhead data that negatively affect these methods in improving resource allocation and anomaly detection in an IoV network. This work introduces three novel models that federate learning and deep reinforcement learning with blockchain technology for guarantees in terms of privacy, scalability, and dynamic decision-making. The first method is Blockchain-Supported Federated Deep Reinforcement Learning (BF-DRL), an integration of the decentralized FL with the ability of DRL in making decisions in matters of optimizing resource allocation as well as secure updates. The second one is Convolutional Recurrent Federated Learning, or CR-FL. CRFL is a design that incorporates the application of Convolutional Neural Networks for feature extraction in space and the incorporation of Long Short-Term Memory for temporal anomaly detection. The last one is adaptive blockchain resource allocation using deep Q-networks. Here, real-time vehicular data samples are used for the optimization of the parameters of the blockchain, which may be found in instances such as transaction speed and block size. This work has also enhanced resource optimization with reduced latency at 25-30% and enhanced attack detection at 99.5% accuracy. Moreover, scalability of blockchain improved the throughput of transactions by 20-30%.